Explore cutting-edge machine learning research in this 58-minute video covering eight papers from NeurIPS 2023. Dive into topics like temporal action segmentation using activity grammars, test-time adaptation of discriminative models with diffusion, noise effects on recurrent neural network learning, sketching algorithms for sparse dictionary learning, equivariant adaptation of large pretrained models, multi-head adapter routing for cross-task generalization, geometry-aware model adaptation, and adversarial learning for feature shift detection and correction. Gain insights into the latest advancements in AI and deep learning from top researchers in the field.
NeurIPS 2023 Poster Session 4 - Highlights in Machine Learning Research
Overview
Syllabus
- Activity Grammars for Temporal Action Segmentation
- Diffusion-TTA: Test-time Adaptation of Discriminative Models via Generative Feedback
- On the Role of Noise in the Sample Complexity of Learning Recurrent Neural Networks: Exponential Gaps for Long Sequences
- Sketching Algorithms for Sparse Dictionary Learning: PTAS and Turnstile Streaming
- Equivariant Adaptation of Large Pretrained Models
- Multi-Head Adapter Routing for Cross-Task Generalization
- Geometry-Aware Adaptation for Pretrained Models
- Adversarial Learning for Feature Shift Detection and Correction
Taught by
Yannic Kilcher
